Abstract In cloud-edge fusion computing, optimizing Virtual Machine (VM) migration is critical to reducing network congestion and energy consumption. This paper proposes three heuristic algorithms—LM, mCaM, and mCaM2—that balance long-term communication efficiency with the immediate overhead of migration. Experimental results demonstrate that these algorithms can reduce total network costs by 20% while operating 70% faster than current state-of-the-art solutions.
Introduction
Cloud-edge fusion computing aims to leverage the high processing power of centralized clouds alongside the low-latency capabilities of edge devices. However, this architecture introduces a significant "network tax," with datacenters often consuming 10–20% of their energy on network traffic alone. Virtual Machine (VM) migration is frequently employed to rebalance loads or reduce latency, but the migration process itself generates a massive burst of traffic. Traditional strategies often overlook this "migration cost," focusing only on the final communication layout. This study addresses this gap by developing a holistic migration strategy that accounts for both immediate and long-term network impacts.
Mathematical Modeling and Methodology
The core of the proposed approach is a mathematical model that defines the total network cost ($Cost\_Total$) as a weighted sum of two primary factors: Communication Cost ($Cost\_Com$) and Migration Cost ($Cost\_Mig$).
- Communication Cost: Represents the ongoing traffic between VMs based on logical dependencies and physical distance.
- Migration Cost: Represents the one-time traffic spike required to transfer a VM's data (its "Size") between physical machines.
Because these costs are measured in different units, the authors utilize a normalization process to map them to a 0–1 range ($Cost\_Com\_std$ and $Cost\_Mig\_std$). This allows for the application of weight coefficients ($\alpha$ and $\beta$) to balance the trade-off between "moving" and "staying."
Proposed Heuristic Algorithms
The study introduces three specific algorithms to solve the optimization problem:
- LM (LocalSearch for min-cost): A greedy approach that evaluates every possible destination for an overloaded VM and selects the one with the lowest total cost.
- mCaM (Min-Communication and Migration Cost): This algorithm introduces the "Network Benefit" metric, prioritizing migrations that offer the greatest improvement over the current state rather than just the lowest absolute cost.
- mCaM2: A sophisticated two-stage variant of mCaM. It first identifies optimal destinations based on communication benefits and then selects VMs to move based on a "benefit-to-cost ratio," preventing the migration of high-traffic VMs if the cost is too high.
Experimental Setup and Results
The algorithms were tested using the BRITE (Boston University Representative Internet Topology Generator) to create realistic network dependencies and a multi-layer tree-structure topology to simulate datacenter architectures.
Key findings include:
- Optimal Weighting: A balance of 70% communication cost ($\alpha = 0.7$) and 30% migration cost ($\beta = 0.3$) provided the best overall network health.
- Performance Gains: The mCaM algorithm achieved a 20% reduction in total network costs compared to the AppAware baseline. While mCaM2 was superior in reducing long-term communication delays, mCaM offered the best overall balance for total network health by effectively managing migration spikes.
- Operational Efficiency: The proposed heuristics were 70% faster than existing solutions, making them suitable for real-time edge computing environments where decisions must be made in milliseconds.
Conclusion
By integrating migration costs into the optimization framework, the LM, mCaM, and mCaM2 algorithms provide a more sustainable approach to VM management in cloud-edge fusion. Future research will focus on implementing these strategies in real-world testbeds and developing dynamic weight coefficients that adapt to real-time network conditions.
Summary updated